DirCast：单流式视频语言模型实现电竞直播导播与解说
原标题：DirCast: Live Esports Directing and Commentary with a Single Streaming Video-Language Model
AI 摘要
来自Alaya Lab、东京大学、佐治亚理工学院、北京大学和东京科学研究所的研究者提出DirCast，一个单一流式视频语言模型，能同时完成电竞直播的导播（每秒决定切换哪个选手镜头）和实时解说，且仅依赖过去信息。团队构建了27场Dota 2职业比赛数据集（含每位英雄镜头、人类导播选择和解说词）。实验显示DirCast导播优于规则和零样本导播（包括GPT-5），其解说在Proact-VL协议下对GPT-4o的比较胜率为29.2%，而11个观看人类直播流的最强流式解说模型为19.4%。
正文节选
1]Alaya Lab 2]The University of Tokyo 3]Georgia Institute of Technology 4]Peking University 5]Institute of Science Tokyo \githubhttps://zixiangao.github.io/DirCast/ \Codehttps://github.com/ZixianGao/DirCast-code \correspondenceKaipeng Zhang DirCast: Live Esports Directing and Commentary with a Single Streaming Video-Language Model Abstract A live esports broadcast needs a director who chooses which camera goes on air and a caster who talks over that choice, every second and without knowing what